Abstract
Biodiversity currently peaks at the equator, decreasing towards the poles. Growing fossil evidence suggest that this hump-shaped latitudinal diversity gradient (LDG) has not been persistent through time, with similar species diversity across latitudes flattening out the LDG during past “greenhouse” periods. This provides a new starting point for LDG research. Most studies assume the processes shaping the LDG have acted constantly through time and seek to understand why diversity accumulated in the Holarctic at lower levels than at the equator, e.g. as the result of limited dispersal, or higher turnover in Holarctic regions. However, fossil investigations suggest that we need to explain when and why diversity was lost at high latitudes to generate the LDG. Unfortunately, diversity lost scenarios in the Holarctic have been repeatedly proposed but not yet clearly demonstrated. Here, we outline the ‘asymmetric gradient of extinction’ (AGE) framework, which contextualize previous ideas behind the LDG in the frame of a time-variable scenario. We suggest the current steep LDG may be explained by the extinction of clades adapted to warmer conditions from the new temperate regions formed in the Neogene, together with the equator-ward dispersal of organisms tracking their own climatic preferences, when tropical biomes became restricted to the equator. Conversely, high rates of speciation and pole-ward dispersal can account for the formation of an ancient flat LDG during the Cretaceous–Paleogene greenhouse period. Phylogenies and fossils of the Testudines, Crocodilia and Lepidosauria support the AGE scenario and showed the LDG to have varied over time, with high latitudes serving as a source of tropical diversity but suffering disproportionate extinction during transitional periods to cold climate. Our results demonstrate that the inclusion of fossils in macroevolutionary studies allows detecting extinction events less detectable in analyses restricted to present-day data only.
Introduction
The current increase in species richness from the poles towards the equator, known as the latitudinal diversity gradient (LDG), is one of the most conspicuous patterns in ecology and evolution. This pattern has been described for microbes, insects, vertebrates, and plants, and for marine, freshwater, and terrestrial ecosystems1–6.
For decades, it has been thought that the modern-type steep LDG (with higher diversity at the equator) persisted throughout the Phanerozoic (the last 540 million years), even if the gradient was sometimes shallower7, based on published fossil record studies8,9. However, the methodological limitations of fossil sampling have called this conclusion into question. Analyses controlling for sampling bias have suggested that, for many groups, the LDG was less marked in the past than it is today, flat (i.e. with similar species diversity across latitudes) or even developed a paleotemperate peak during some periods in the past (see Mannion et al., (2014) for a review). This sampling-corrected flatter LDG in deep time has been demonstrated for non-avian dinosaurs11, mammals12,13, tetrapods14, insects15–17, brachiopods18–20, bivalves21, coral reefs22, crocodiles23, turtles24,25, and plants26–28. The pattern emerging from fossil studies also suggests that steep LDGs, such as that currently observed, have been restricted to the relatively small number of short “coldhouse” periods during the history of the Earth: the Ordovician/Silurian, the Carboniferous/Permian, the end of the Jurassic, and the Neogene. Most of the Phanerozoic has instead been characterized by warm greenhouse climates associated with a flatter LDG1 (Fig. 1).
This recent fossil evidence now provides a new starting point for LDG research. Most hypotheses on the origin of the LDG are based on the assumptions that (i) the processes shaping the LDG have acted constantly through time and (ii) equatorial regions are the source of world diversity29,30, with the LDG resulting from lower levels of diversity accumulation in the Holarctic than at the equator through time7,31,32. Previous studies have explained the LDG as a result of greater tropical diversification and limited dispersal out of the equatorial region7,31,33, or by high rates of turnover in the Holarctic (i.e. similar high speciation (λ) and extinction (μ) rates; λ ≈ μ), all keeping diversity levels in this region low over time (Table 1), for amphibians34,35, birds32,36,37, butterflies38, conifers39, fishes40, mammals33,37, and lepidosaurs41, for example. However, the recent fossil investigations showing, for many lineages, similar high diversity levels in the Holarctic and at the equator in the past suggest we do not necessarily need to explain why diversity accumulated at slower rates in the Holarctic through time, but the question being how and when diversity was lost at high latitudes, giving rise to the current shape of the LDG10?
Diversity losses in the Holarctic have been traditionally considered to underlie the LDG37. They were initially attributed to Pleistocene glaciations42, but this hypothesis has been called into question by the finding that the LDG substantially predates the Pleistocene7. More ancient extinctions have also been considered43–49. For example, Hawkins et al. (2006)46 suggested that the avian LDG resulted from the differential extirpation of older warm-adapted clades from the temperate regions newly formed in the Neogene. Pyron (2014)41 suggested that higher temperate extinction represents a dominant force for the origin and maintenance of latitudinal gradients. More recently, Pulido-Santacruz & Weir (2016)49 proposed the terrestrial LDG is largely the effect of a post-Eocene increase in extinction rates at high latitudes resulting from the cooling Cenozoic trend. Unfortunately, using phylogenies alone, these and other studies on the LDG have not clearly demonstrated diversity losses in the Holarctic but instead high regional turnover 33,35–39,50. Nonetheless, high turnover can only explain a slow accumulation of lineages, with one fauna being replaced by another, but does not explain diversity decline (i.e. a reduction in the net number of species). Diversity declines occur when extinction exceeds speciation, resulting in negative net diversification rates (r = λ − μ; r < 0). Accordingly, ‘diversity loss’ hypotheses need to be differentiated in the literature from ‘high turnover’ scenarios. The perceived difficulty for inferring negative diversification rates from present-day phylogenetic data51,52 and the assumption that diversity levels were always lower in the Holarctic than at the equator have resulted in ‘diversity loss’ hypotheses being repeatedly proposed but seldom demonstrated. Meanwhile, numerous fossil investigations have detected signatures of extinction and diversity loss in the Northern Hemisphere. For instance, Archibald et al.15,16 sampled insect diversity at an Eocene site in Canada, and in present-day temperate Massachusetts (USA) and tropical sites of Costa Rica. Insect diversity was higher at the Eocene paleotropical site than the modern temperate locality, and comparable to the modern-day tropical locality, suggesting that post-Eocene insects have thus suffered greater levels of extinction in the Nearctic regions than around the equator. This pattern is consistent with results of other studies on various taxonomic groups, including birds53, invertebrates15,16,54, mammals12,13,55 and plants 56–58. However, fossil studies are generally restricted to a geographic and temporal scale, which makes difficult to extrapolate local inferences of extinction in the context of the LDG.
Here we capitalize on the current knowledge and we aim at: (1) outlining a new framework for studying the LDG, the ‘asymmetric gradient of extinction’ (AGE) hypothesis, which formalizes and expands previous hypotheses about the contribution of extinctions to the LDG in the context of a time-variable pattern; and (2) testing the predictions of the AGE hypothesis with three tetrapod groups.
From a flat to a hump-shaped LDG: an asymmetric gradient of extinction
Over the Earth’s history, the geographic extent of the tropical biome around the equator has fluctuated, with periods of pole-ward expansion during which warm paratropical conditions appeared at high latitudes, followed by periods of equator-ward contractions59–63 (Fig. 1). To account for these biodiversity changes, we propose to include a temporal component to study the LDG in which prevailing speciation, extinction and dispersal dynamics change between warm and cold intervals. The AGE hypothesis makes the following main postulates (P) during climatic transitions towards cooler climates (Fig. 2; Table 1): (P1) extinction rate exceeds the speciation rate at high latitudes (H), i.e. declining diversity (rH < 0), while the equatorial (E) regions remain stable, and (P2) the rate of dispersal ‘into the equator’ is greater than that ‘out of the equator’ (dHE > dEH), thus triggering the formation of a steep LDG. Conversely, during the transition to greenhouse periods, different processes drove the formation of a flat gradient, and AGE postulates: (P3) diversification rates increase in the new paratropical regions (rE < rH), and (P4) ‘out of the tropical’ dispersals increase over ‘into the equator’ migrations (dHE < dEH).
For instance, the last 100 million have witnessed the contraction of tropical conditions towards the equator, due to the global cooling since the latest Cretaceous–early Cenozoic period (the most recent greenhouse period), culminating in the Pleistocene glaciations64. According to the AGE hypothesis, the expansion of tropical-like conditions to higher latitudes induced species diversification in the new paratropical areas (Fig. 2a) and facilitated movements within the broad ‘tropical belt’, such that tropical equatorial clades were able to disperse ‘out of the equator’ into high-latitude warm regions29,65. An equable Cretaceous-early Cenozoic greenhouse climate thus triggered the formation of a flat LDG (Fig. 2a). By contrast, tropical biome retractions following the climate cooling induced periods of declining diversity at high latitudes (where climate change was more intensively felt), and initiated biotic movements ‘into the equator’ (Fig. 2b). Extinction rates were high for tropical-adapted lineages at high latitudes, but lower for low-latitude tropical lineages. Climate change would thus have driven the development of an asymmetric gradient of extinction (AGE) and dispersal within the tropical biome, and mediated the formation of a steep LDG (Fig. 2c).
The AGE hypothesis attempts to reconcile previous contending hypotheses on the origin of the LDG by placing them in a temporal scenario (Table 1). For instance, there is controversial support around the tropics being ‘cradle’ or ‘museum of diversity’ 66, and dispersal prevailing ‘out of’29,65 or ‘into the tropics’38,41,50. The AGE hypothesis alternatively invokes the ‘museum of diversity’ regarding the equatorial tropics as refuge during coldhouse transitions, but also the ‘cradle of diversity’ during greenhouse periods. Similarly, the hypothesis invokes ‘out of the equator’ dispersals during greenhouse transitions and ‘into the equator’ dispersals during coldhouse transitions. The AGE hypothesis also assumes tropical niche conservatism due to physiological limits30,31 as we postulate that most of the tropical-adapted taxa at high latitudes were unable to adapt and either went extinct or suffered restrictions of their distributions when the tropical biome retreated towards the equator. Accordingly, the AGE hypothesis focuses on the fate of tropical-adapted clades under variable climate conditions but does not apply for groups having evolved the appropriate adaptations to cope with climate change, e.g.67,68.
Here, we used comparative methods for both phylogeny-based and fossil-based datasets to test the predictions of the AGE hypothesis for the Testudines, Crocodilia and Lepidosauria. The modern-day Crocodilia and Lepidosauria comprise mostly tropical-adapted species with a classic LDG pattern as shown by an accumulation of diversity at equatorial latitudes41,44. We evaluated the applicability of our framework to subtropical taxa as well, by extending the study to Testudines, a group displaying a hump-shaped gradient of diversity today centred on subtropical latitudes (10°S–30°N)69. By contrast, the paleolatitudinal distribution of turtles was concentrated in the Holarctic (30–60°N) during the Cretaceous24,25. All these lineages are ancient and likely experienced climatic transitions during the early Cenozoic23,25,41,44,69. They display contrasting patterns of species richness: turtles and crocodiles are species-poor (330 and 25 species, respectively), while lepidosaurs include a large number of species (9500+ species) and have a rich fossil record extending back to the Triassic (Early Cretaceous for crocodiles), providing information about the variation of latitudinal species richness accumulation during their evolutionary history.
Results
Phylogeny-based diversification analyses: are diversification rates higher at the equator?
According to current distribution data, the species richness of turtles, lepidosaurs and crocodiles peaks near the equator, with 84% of all extant species living in the tropics, only 15% living in temperate regions and 1% spanning both biomes. We classified each species reported in the phylogeny (Supplementary Tables 1-3) as living close to the equator (the modern-day tropical biome) or the Holarctic and Southern Hemisphere (the modern-day temperate biome). For turtles, there were 239 tropical species, 84 temperate and 6 spanning both biomes (7 were marine species). For lepidosaurs, there were 7955 tropical species, 1337 temperate and 124 spanning both biomes. The species-poor crocodile clade had only 23 tropical and two temperate species.
We analyzed differences in diversification rates between the Holarctic and equatorial regions, with the binary state change speciation and extinction model ‘BiSSE’70,71, see Methods. We did not use the geographic state change speciation and extinction model72, which is appropriate for dealing with widespread species, because most of the species in our datasets were endemic to the Holarctic or equatorial regions, and, for a character state to be considered in SSE models, it must account for at least 10% of the total diversity73. We did not apply the BiSSE model to crocodiles, because simulation studies have shown that trees containing fewer than 300 species may have to weak a phylogenetic signal to generate sufficient statistical power73.
We first used the time-constant BiSSE model, which is generally used in studies of the LDG. For turtles, net diversification rates were higher in the Holarctic than at the equator (Table 2, Supplementary Fig. 1a), but this difference was not significant, and rates of dispersal ‘into the equator’ were ten times higher than those ‘out of the equator’. For lepidosaurs, a similar dispersal pattern was recovered, but net diversification rates were significantly higher towards the equator (Supplementary Fig. 1b). We tested the AGE hypothesis by introducing two shift times, at 51 and 23 million years ago (Ma), to detect differences in diversification dynamics between greenhouse, transitional, and coldhouse periods. This model indicated that the net diversification of turtles was similar in the Holarctic and at the equator, whereas it was lower in the Holarctic for lepidosaurs until the coldhouse period, when Holarctic diversification increased (Table 2, Supplementary Fig. 2). Dispersal was considered to be symmetric between regions (into the equator = out of the equator) during greenhouse periods, and asymmetric (into the equator > out of the equator) during the climatic transition and coldhouse period. The same patterns were obtained for analyses with the same model but with different combinations of shift times (51/66 Ma and 34/23 Ma; Supplementary Fig. 3).
Fossil-based diversification analyses: evidence for ancient tropical extinctions?
We also analyzed differences in diversification rates between the Holarctic and equatorial regions based exclusively on fossil data. The turtle fossil dataset comprised 4084 occurrences for 420 genera (65 extant and 355 extinct; Supplementary Table 4). The lepidosaur fossil dataset comprised 4798 occurrences for 638 genera (120 extant and 518 extinct; Supplementary Table 5). The crocodile fossil dataset comprised 1596 occurrences for 121 genera (9 extant and 112 extinct; Supplementary Table 6). We first inferred global diversification dynamics by analyzing the fossil datasets as a whole with a Bayesian approach to inferring the temporal dynamics of origination and extinction rates based on fossil occurrences74 (see Methods). For turtles, origination rates peaked during the Jurassic, subsequently decreasing until the present day. Extinction rates were generally low and constant during the Mesozoic, but increased during the coldhouse periods of the Jurassic and Paleogene, resulting in negative net diversification during the Paleogene (Fig. 3, Table 2, Supplementary Figs. 4, 5). For lepidosaurs, origination rates peaked in the Jurassic and Late Cretaceous, whereas extinction increased steadily until the Late Cretaceous. In the Paleogene, net diversification approached zero, suggesting a high rate of turnover (Fig. 3, Supplementary Figs. 6, 7). Crocodile origination peaked in the Early Cretaceous, subsequently decreasing towards the present day, and extinction rates were generally low and constant. We also identified diversity losses in the Paleogene extending to the present, suggesting that crocodiles are still in a phase of declining diversity (Fig. 3, Supplementary Figs. 8, 9).
We performed additional analyses with different subsets of the three fossil datasets, to separate the speciation and extinction signals between geographic regions (equator or Holarctic) and ecological conditions (temperate or tropical, see Methods). These analyses showed that the diversity losses experienced by turtles and crocodiles during the Paleogene were mostly attributable to species living in the Holarctic and under tropical conditions (Figs. 4, 5, Table 2). The global diversity loss inferred for crocodiles during the Neogene was attributed to taxa living in both the Holarctic and equatorial regions (adapted to temperate and tropical conditions respectively), providing further support for the hypothesis that this whole group is in decline. For all groups, temperate taxa have been estimated to have high rates of diversification during the Oligocene, but lower rates during the Neogene. For the equatorial datasets, extinction and speciation rates decreased over time, resulting in constant net diversification rates (except for lepidosaurs, which displayed a decrease in diversification during the Paleogene, followed by an increase in diversification during the Neogene).
Estimations of ancestral origins: did groups preferentially originate close to the equator?
We tested the predictions of the AGE hypothesis further, by performing biogeographic analyses with the dispersal-extinction-cladogenesis (DEC) model75 and dated phylogenies (see Methods). We first analyzed the data in an unconstrained DEC analysis in which all ranges covering three areas could be in an ancestral state. We inferred an equatorial distribution for the deepest nodes for the turtles and lepidosaurs, whence these lineages colonized the other regions (Fig. 6a, Table 2, Supplementary Fig. 10). Crocodile ancestors were found to have been widespread during the Cretaceous, with an early vicariant speciation event separating Alligator in the Holarctic from the other Alligatoridae genera in equatorial regions (Supplementary Fig. 11).
Our biogeographic estimates based exclusively on extant data conflict with the fossil record23,24,76. We overcame this bias by introducing information about the distribution of fossils into DEC, in the form of hard (HFC) and soft (SFC) geographic fossil constraints at specific nodes (see Methods; Supplementary Tables 7–9). The inclusion of fossil information yielded very different biogeographic histories for the three groups (Table 2; turtles: Fig. 6b, Supplementary Fig. 12; lepidosaurs: Supplementary Figs. 13, 14; and crocodiles: Supplementary Figs. 15, 16). Under the SFC model, turtles were found to have originated in the Northern Hemisphere (under the HFC model they were spread over both regions), whence lineages migrated towards the equator and southern regions (Fig. 6b, Supplementary Fig. 12). Most dispersal therefore occurred ‘into the equator’ (Supplementary Fig. 17, Supplementary Table 10). We also detected a larger number of geographic extinctions when fossil ranges were considered, predominantly for turtle lineages in the Holarctic (53 and 11 lineages disappeared from this region under the HFC and SFC models, respectively) and in southern temperate regions (9 in the HFC model; Supplementary Fig. 17, Supplementary Table 11). The same trend was observed when the number of extinction/dispersal events was controlled for the number of lineages currently distributed in each region (Fig. 7).
Lepidosaurs originated in both regions in both SFC and HFC analyses (Supplementary Figs. 13, 14). During the greenhouse period, dispersal ‘into the equator’ occurred at the same rate (or at a higher rate in the HFC model) than dispersal ‘out of the equator’, and dispersal ‘out of the equator’ prevailed thereafter (Supplementary Fig. 17, Supplementary Table 10). Estimated range extinction rates were high in this group under the unconstrained model, with 30 lineages extirpated from the Holarctic, two from southern temperate regions and 152 from the equator (Supplementary Fig. 17, Supplementary Table 11). Under fossil-informed models, the number of Holarctic extinctions was higher (109 and 66 lineages in the HFC and SFC models, respectively), whereas the number of lineages extirpated from the equator was similar (144 and 109 in the HFC and SFC models, respectively; Supplementary Fig. 17). When the number of events was controlled for the actual number of lineages distributed in each region, the number of Holarctic extinctions and dispersals ‘into the equator’ increased dramatically, exceeding equatorial dispersal/extinctions (Fig. 7). For crocodiles, analyses including fossil ranges showed that all the early nodes were distributed in the Holarctic (Supplementary Figs. 15, 16), and range extinctions were detected: four lineages disappeared from the Holarctic, three from southern temperate regions, and two from the equator (HFC model; Supplementary Fig. 17, Supplementary Tables 11, 12). Only two lineages disappeared from the Holarctic in the SFC model. The same trends were observed after controlling the number of events for the current number of lineages in each region (Fig. 7).
Discussion
Generation of the current LDG
Fossil investigations have shown that at certain times during the Phanerozoic, the LDG has weakened, flattened, or developed a palaeotemperate peak, with diversity at high latitudes being greater in some periods of the past than currently for many groups10,13. Hypotheses relating to ‘slow Holarctic diversity accumulation’, such as limited dispersal to the Holarctic31, high Holarctic turnover37,41,49, or high rates of equatorial diversification 32–35,77, cannot themselves account for the formation of a flatten LDG, or the transition from higher to lower diversity in the Holarctic observed in many groups. Furthermore, although the processes shaping biodiversity vary over time and space, this has been largely overlooked in the context of the LDG, which has been generally explained in terms of the actions of time-constant process.
We account for temporal changes in the global distribution of biodiversity by proposing a scenario involving losses/gains of tropical diversity at high latitudes during transitional periods from warm to cool conditions. The AGE hypothesis captures components of previous studies10,31,35,50,65 in the context of a time-variable LDG to disentangle the relative contributions of speciation, extinction and dispersal for each particular geological period in the formation of the LDG. Below, we evaluate the support of our analyses to the postulates of the time-variable AGE hypothesis (Fig. 2; Table 1):
P1: Extinction exceeds speciation at high latitudes during cool transitions, i.e. declining diversity (rH < 0)
Our diversification analyses based on extant species (time-constant and time-variable BiSSE analyses) do not support this postulate, suggesting instead higher levels of Holarctic diversification for turtles, and of equatorial diversification for lepidosaurs. By contrast, results for fossil-only (PyRate) diversification analyses were consistent with this prediction for turtles and crocodiles. We found that diversification rates of turtles and crocodiles decreased in all regions during the transition to colder climates, but the slowing of diversification was much stronger in the Holarctic than at the equator, with extinction exceeding speciation in this region (Fig. 3, 4). This suggests that Holarctic diversity loss (r < 0) during the Paleogene could explain the formation of a steep LDG for these groups. For lepidosaurs, P1 is not supported by fossil data; diversity losses occurred during Cenozoic cooling in the equator but not at higher latitudes (Fig. 4). Diversity dynamics for the species distributed at the equator, however, may not be entirely reliable, due to the poverty of the equatorial dataset in terms of the number of fossil lineages and the small number of records per lineage (Supplementary Table 12). Uncertainties therefore remain on these estimates, which have wide credibility intervals, probably due to geographic biases in the fossil record78. Turnover rates were very high in the Holarctic during the transitional period, indicating that species did disappear from high latitudes, but that a new lepidosaur community replaced them. This result suggests the number of lepidosaur species may always have been unbalanced between regions. The high Holarctic turnover would contribute to the maintenance of this pattern, together with the inferred temporal increases in diversification at the equator (Fig. 4), as previously hypothesized 41.
P2: Higher ‘into the equator’ dispersal than ‘out of the equator’ dispersal (dHE > dEH) during cool transitions
The DEC biogeographic analyses based on extant species do not support this postulate but instead the ‘tropical niche conservatism’ hypothesis for turtles, with an equatorial origin and recent invasion of high-latitude regions, resulting in less time for lineages to diversify31 (Figs. 6, 7, Table 1, 2). This result is consistent with the findings of recent investigations79,80. For lepidosaurs, they support the ‘out of the tropics’ (Fig. 7, Supplementary Figs. 1-3, 10), and for crocodiles the diversification hypothesis, with higher origination rates close to the equator and no effect on dispersal (Fig. 7, Table 2, Supplementary Fig. 11). In contrast, time-constant and time-variable BiSSE analyses are consistent with P2 and the ‘into the equator’ hypothesis to explain the LDG (Supplementary Figs. 1a, 2a, 3a), as so they are the results for fossil-informed DEC biogeographic analyses; all groups are inferred to have had a widespread ancestral distribution that subsequently contracted towards the equator due to both higher levels of range extirpations at higher latitudes and ‘into the equator’ dispersals during Cenozoic cooling (Figs. 6, 7). This result is also in agreement with previous fossil investigations on turtles 24,25,76 and crocodiles 23,44. For lepidosaurs, in absolute terms, more species migrated “out of” than “into the equator” (Supplementary Fig. 17), but the number of species in the equatorial region today is four times the number of lineages elsewhere. After controlling for the imbalance in species sampling in our tree, we found that a higher proportion of lepidosaur species lost their ancestral Holarctic distribution and emigrated ‘into the equator’ 41 than the other way around (Fig. 7). Although the number of fossil constraints in the biogeographic analysis of lepidosaurs was relatively low given the size of the tree (30 Holarctic and equatorial fossils for 4161 nodes), these constraints significantly increased the absolute number of Holarctic range extinctions (from 30 to 109) and ‘into the equator’ dispersals (from 40 to 124) relative to estimates without such constraints (Supplementary Tables 10, 11). Meanwhile, the inclusion of fossil data did not alter the number of events estimated for equatorial taxa. This finding suggests that a deeper understanding of lepidosaur fossil taxonomy might facilitate the assignment of fossils on the tree, and the detection of additional high-latitude extinctions not detected here.
Unfortunately, the age of the taxa evaluated here prevents us for testing the predictions associated with the transition from coldhouse to greenhouse conditions (P3-4). Nonetheless, our fossil-based analyses show similar diversification rates in the Holarctic and equatorial regions during the greenhouse period of the Cretaceous-early Cenozoic for all groups (overlapping credibility intervals; Fig. 4; Supplementary Figs. 1-3, 4-9), consistent with the idea of the existence of a flattened LDG during this phase 10. Similarly, the AGE hypothesis focuses essentially on the Northern Hemisphere, but diversity losses and dispersals ‘into the equator’ may have also occurred in the temperate regions of the Southern Hemisphere. Indeed, we found high rates of range extinctions in this region for all groups (P2; red lines on Fig. 7). Unfortunately, the scarce fossil record prevents any diversification estimates for this region (see Methods).
Overall, the general pattern that could be extracted from our study is that the AGE hypothesis was supported for crocodiles and turtles using fossil and fossil-informed phylogenetic investigations (Table 1, 2; P1–P2 of AGE). However, if we rely only on analyses based on data for extant species using biogeographic and constant and time-variable BiSSE diversification models, this evolutionary scenario was poorly supported (although BiSSE analyses support P2 of AGE). Support for the AGE hypothesis is mixed in lepidosaurs; on the one hand the detected Holarctic range contractions are in agreement with “higher Holarctic diversity loss” scenarios (P2 of AGE), on the other hand, the evidence for high Holarctic turnover are more in line with previous “slower Holarctic diversity accumulation” hypotheses.
The timing and effect of the last greenhouse to coldhouse climatic transition
Recent fossil investigations suggest that changes in the shape of the LDG have been associated with major climatic oscillations10,13. Accordingly, we hypothesized the impoverishment of the Holarctic resulted from the contraction of the tropical biome during the last greenhouse to coldhouse transition. But when this coldhouse transition took place? The transitional period to cold was here defined between 51 and 23 Ma, after the early Eocene Climatic Optimum (EECO), based on paleontological evidence showing that paratropical conditions and the associated warm-adapted taxa disappeared from high latitudes between the mid-late Eocene and the Neogene59,61,62. Our diversification results with time intervals defined by the main climatic periods are consistent with these observations, and detect Holarctic diversity loss during the late Paleogene. We cannot exclude, however, that diversity losses at high latitudes occurred since the Cretaceous, as suggested by our fossil-based diversification analyses with time intervals defined by the main geological periods (Supplementary Figs. 4, 6, 8) and by our fossil-based biogeographic analyses suggesting the prevalence of ‘into the equator’ dispersals since the Cretaceous (Fig. 7, Supplementary Table 11). These findings could suggest that other processes different to climate change mediated the extinction and range contraction of Holarctic lineages in the Cretaceous, or alternatively, that a transition phase to cold started before the relatively short interval considered here. Some studies consider the EECO only represented a transient temperature peak within an otherwise cooling trend that started in the Cretaceous64,81. This trend was intensified by the Cretaceous-Paleogene (K-Pg) mass extinction, and the drop in temperatures caused by the impact-associated winter82,83. In our study, lineage extinctions, range extinctions and southward dispersals increased between the K-Pg and Neogene (Fig. 4, 7), suggesting an additive effect of K-Pg and Neogene cooling on depopulation of the Holarctic.
The ancestors of turtles, lepidosaurs and crocodiles were adapted to tropical conditions during the Late Cretaceous44,84. Our results indicate that extinction events were not random, instead preferentially affecting taxa living in tropical-like climates at high latitudes48 (Figs. 4, 5). This suggests that many species adapted to warm conditions living in the Holarctic were unable to adapt to the new temperate regimes and either went extinct or escaped extinction by contracting their ranges in a southerly direction (Fig. 7). Meanwhile, we found that the diversification rates of turtles, crocodiles and lepidosaurs living in temperate climatic conditions were significantly higher than those of tropical-adapted taxa living in Holarctic and equatorial regions after the transition to temperate climates in the late Eocene (Fig. 5, Table 2). The new temperate habitats could have constituted an opportunity for diversification because they increased geographic ranges and ecological niches34, and may have driven an inverse LDG for some groups39,68. Several radiations following the appearance of the temperate biome have been identified in other groups of organisms, such as plants 85,86, mammals 87,88 or insects89. After this period, speciation decreased dramatically in the temperate lineages of our focal groups, possibly due to the effect of the Pleistocene glaciations, and no difference in diversification between tropical and temperate lineages is currently evident (Fig. 5). In summary, our study suggests that differences in species richness between geographic regions may be explained by differences in diversification and dispersal rates. Differences in species richness between ecological types may be explained by the longer time available for tropical-adapted clades to diversify in tropical areas90 rather than higher rate of speciation under warm tropical environments, as previously postulated 91.
Reconciling fossil and phylogenetic evidence
Our results unequivocally demonstrate that the inclusion of fossils in macroevolutionary studies makes it possible to detect signals of ancient high-latitude extinctions and range extirpations (Figs. 4-7), otherwise hardly detectable with analyses based exclusively on present-day data. This conflict between extant and fossil evidence may extend beyond our study, pervading the LDG literature. High extinction rates have occasionally been inferred in tropical lineages67,92–94, with hypotheses relating to extinction focused on temperate taxa and recent time scales, such as the effects of recent Pleistocene glaciations, for example33,37,41,42. In reported cases of extinction, origination rates were also found to be elevated in high-latitude groups (high turnover) 33,37–39,41,50, while diversity losses (r < 0) have to our knowledge never been inferred in phylogenetic studies of the LDG (with the exception of the recent Pulido-Santacruz & Weir (2016) study using time constant BiSSE models, but see below). On the other hand, ancient tropical extinction at high latitudes is supported by fossil studies on various taxonomic groups15,16,28,53–56.
The last decade has seen many efforts to reconcile fossil and phylogenetic evidence. Birth-death diversification models have been developed to detect negative diversification rates on reconstructed phylogenies, or total-evidence trees95–97. Still, their use in the literature is limited and these models are difficult to implement in a trait evolution context, such as in the study of the LDG. LDG studies are often based on state-dependent speciation and extinction models35,38,40,41,49,50. These models are designed to test differential diversification and asymmetric transition scenarios, such as that suggested here, but LDG studies often assume that diversification parameters remain constant over time. If the evolutionary processes shaping the LDG have varied across latitudes and time, then time-constant models are not appropriate for testing more complex scenarios underlying the LDG. Moreover, the potential of time-constant models for detecting negative diversification rates is questionable, since inferring negative diversification for the entire history of lineages conflicts with the fact that these groups are still extant. Testing our hypothesis thus requires the implementation of time-variable models. When applied to the study of diversity patterns, these models have revealed marked extinction signatures in ancestral tropical plant clades98. The incorporation of time-shifts into our BiSSE analyses improves but not completely reconciles the fossil evidence with extant diversity. Identifying the causes of this problem and finding solutions are beyond the scope of this study, but this artifact highlights the importance of fossils in macroevolutionary inferences99. Fossil records remain incomplete, but they nevertheless provide the only direct evidence of the diversity that existed in the past. By contrast to molecular phylogenies, the incompleteness of the fossil record has a less problematic effect on the estimation of speciation and extinction rates, because removing a random set of taxa does not affect the observed occurrences of other lineages74. Indeed, simulations have shown that PyRate correctly estimates the dynamics of speciation and extinction rates under low levels of preservation or severely incomplete taxon sampling.
Conclusion
After decades of research, the processes shaping the LDG remain among the most hotly debated topics in ecology and evolutionary biology. We propose here the AGE hypothesis, which explains the origin of the current LDG through the changes in global diversification and dispersal dynamics imposed by large-scale climatic transitions. Our analyses for turtles and crocodilians indicated that the processes shaping the LDG have changed over time, the current form of this gradient being the result of ancient high-latitude tropical diversity loss and range contractions as a consequence of the retraction of the tropical biome and due to climate cooling. The AGE hypothesis might account for the LDG of tropical-adapted groups that were once diverse at high latitudes, but might not be fully applicable to all organisms currently displaying a LDG, as shown here for lepidosaurs.
Methods
Time-calibrated phylogenies and the fossil record
We compared the predictions of the AGE hypothesis with the LDG of three vertebrate groups: turtles (order Testudines), crocodiles (order Crocodilia), and scaled lizards (order Lepidosauria). A time-calibrated phylogeny for each group was obtained from published data. For turtles, we used the phylogeny of Jaffe et al. (2011), including 233 species. We preferred this phylogeny over other more recent and slightly better sampled trees101 because the divergence time estimates of Jaffe et al. (2011) are more consistent with recent estimates based on genomic datasets79,102. For lepidosaurs, we retrieved the most comprehensive dated tree available, including 4161 species41, and a complete phylogeny was obtained for crocodiles103.
Fossil occurrences were downloaded from the Paleobiology Database (https://paleobiodb.org/#/, last accessed October 25th 2017). We reduced potential biases in the taxonomic assignation of turtle, crocodile and lepidosaur fossils, by compiling occurrence data at the genus level. The fossil datasets were cleaned by checking for synonymies between taxa and for assignment to a particular genus or family on the basis of published results (Supplementary Table 4–6).
Estimation of origination and extinction rates with phylogenies
We investigated possible differences between Holarctic and equatorial regions, by combining the turtle and lepidosaur phylogenies with distributional data (Supplementary Tables 1, 2) to fit trait-dependent diversification models in BiSSE70. We accounted for incomplete taxon sampling in the form of trait-specific global sampling fraction of extant species104.
We ensured comparability with previous LDG studies, by initially using a constant-rate trait-dependent diversification model. The constant-rate BiSSE model has six parameters: two speciation rates (without range shift, or in situ speciation), one associated with the Holarctic (hereafter ‘H’, λH) and the other with other equatorial and subtropical regions (hereafter ‘equator’ or ‘E’, λE), two extinction rates associated with the Holarctic (μH) and the equator (μE), and two transition rates (dispersal or range shift), one for the Holarctic to equator direction (qH-E), and the other for the equator to Holarctic direction (qE-H).
We then assessed the effect of species distribution on diversification, allowing for rate changes at specific time points. This approach is associated with a lower bias than the use of constant rates. We used the time-dependent BiSSE (BiSSE.td) model, in which speciation, extinction, and dispersal rates are allowed to vary between regions and to change after the shift times. We introduced two shift times to model different diversification dynamics between greenhouse, transitional, and coldhouse periods. We assumed that a global warm tropical-like climate dominated the world from the origin of the clades until 51 Ma (corresponding to the temperature peak in the Cenozoic). Thereafter, the climate progressively cooled until 23 Ma (the transitional period), when the climate definitively shifted to a temperate-like biome in the Holarctic61,62,64. The shift times at 51 Ma and at 23 Ma are initial values that are re-estimated by the model during the likelihood calculation. The climatic transition in the Cenozoic may have different temporal boundaries, with potential effects on the results. We thus applied the same model but with different combinations of shift times (we tested 51/66 Ma and 34/23 Ma for the upper and lower bounds of the climatic transition).
Analyses were performed with the R package diversitree 0.9-771, using the make.bisse function to construct likelihood functions for each model from the data, and the functions constrain and find.mle to apply different diversification scenarios. Finally, we used a Markov Chain Monte Carlo (MCMC) approach to investigate the credibility intervals of the parameter estimates. Following previous recommendations71, we used an exponential prior 1/(2r) and initiated the chain with the parameters obtained by maximum likelihood methods. We ran 10,000 MCMC steps, with a burn-in of 10%.
Estimation of origination and extinction rates with fossils
We also used fossil data to estimate diversification rates over time. We analyzed the three fossil records, using a Bayesian model for simultaneous inference of the temporal dynamics of origination and extinction, and of preservation rates74. This approach, implemented in PyRate105, uses fossil occurrences that can be assigned to a taxon, in this case fossil genera. The preservation process is used to infer the individual origination and extinction times of each taxon from all fossil occurrences and an estimated preservation rate; it is expressed as expected occurrences per taxon per million years.
We followed a birth-death shift approach106, which focuses on the variation of origination and extinction at a global scale and over large temporal ranges. We used a homogeneous Poisson process of preservation (-mHPP option). We also accounted for the variation of preservation rates across taxa, using a Gamma model with gamma-distributed rate heterogeneity (-mG option). We used four rate categories to discretize the gamma distribution, to allow for a greater variability of preservation rates across taxa.
Given the large number of occurrences analyzed and the vast timescale considered, we dissected the birth–death process into time intervals, and estimated origination and extinction rates within these intervals. In one set of analyses we defined the time intervals using the geological epochs of the stratigraphic timescale107 (Supplementary Figs. 4, 6, 8). In another set of analyses, we defined the intervals according to the different climatic periods characterizing the Cenozoic (Supplementary Figs. 5, 7, 9), as discussed above: the greenhouse world (Cretaceous), the climatic transition (Paleogene), and the coldhouse world (Neogene until the present). We adopted this solution as an alternative to the algorithms implemented in the original PyRate software for joint estimation of the number of rate shifts and the times at which origination and extinction shift74. The estimation of origination and extinction rates within fixed time intervals improved the mixing of the MCMC and made it possible to obtain an overview of the general trends in rate variation over a long timescale 106. Both the preservation and birth–death processes were modeled in continuous time but without being based on boundary crossings. Thus, the origination and extinction rates were measured as the expected number of origination and extinction events per lineage per million years. One potential problem when fixing the number of rate shifts a priori is over-parameterization. We overcame this problem by assuming that the rates of origination and extinction belonged to two families of parameters following a common prior distribution, with parameters estimated from the data with hyper-priors108.
We ran PyRate for 10 million MCMC generations on each of the 10 randomly replicated datasets. We monitored chain mixing and effective sample sizes by examining the log files in Tracer 1.6109. After excluding the first 20% of the samples as a burn-in, we combined the posterior estimates of the origination and extinction rates across all replicates to generate plots of the change in rate over time. The rates of two adjacent intervals were considered significantly different if the mean of one lay outside the 95% credibility interval of the other, and vice versa. We looked at the marginal posterior distributions of origination and extinction rates through the evolutionary history of the three groups and assessed the effect of different environments.
In the context of the LDG, we performed additional analyses with different subsets of fossils, to separate the speciation and extinction signals of different geographic regions (equator or Holarctic) and ecological conditions (temperate or tropical). For example, for turtles, we split the global fossil dataset into four subsets: one for the fossil genera occurring at the equator (429 occurrences), one for the fossils occurring in the Holarctic (3568 occurrences), one for the fossil genera considered to be adapted to temperate conditions (993 occurrences), and one for the fossils considered to be adapted to tropical conditions (2996 occurrences). We excluded the few fossil occurrences for the southern regions of the South Hemisphere (about 180) only in subset analyses, as they were poorly represented in our dataset. Note that a given fossil can be present in both the ‘Holarctic’ and ‘tropical’ datasets. We encoded tropical/temperate preferences by considering macroconditions in the Holarctic to be paratropical until the end of the Eocene, as previously reported 61,62 (and references therein). We also assumed that taxa inhabiting the warm Holarctic were adapted to tropical-like conditions (i.e. a high global temperature, indicating probable adaptation to tropical climates). This is, of course, an oversimplification that may introduce bias into the analysis, but general patterns may nevertheless emerge from such analyses110. For turtles, this assumption is supported by a recent study modeling the climatic niche of this group during the Late Cretaceous, which found that the Holarctic ancestors of turtles were adapted to tropical conditions84. After the late Eocene, we categorized each species as living in the temperate biome or the tropical biome, according to the threshold latitudes defining the tropics (23.4°N and 23.4°S) suggested in a previous study33. This delineation is also consistent overall with the Köppen climate classification. With these datasets, we reproduced the same PyRate analyses as for the whole dataset (see above). In general, the fossil datasets included mostly Holarctic fossils, with a smaller number of occurrences for the equator. Caution is therefore required when drawing conclusions from the equatorial datasets.
Inferring ancestral geographic distribution with phylogenies and fossils
We performed biogeographic analyses with the parametric likelihood method DEC75 using the fast C++ version111 (https://github.com/rhr/lagrange-cpp). Turtle, lepidosaur, and crocodile species distributions were obtained from online databases (www.iucnredlist.org and www.reptile-database.org). We chose 23.4°N and 23.4°S as the threshold latitudes defining the tropics, and categorized each species as living in the Holarctic, in the southern temperate regions, or in the equatorial tropics and subtropical regions. We considered that all ranges comprising three areas could be considered an ancestral state (maxareas =3).
We set up three different DEC analyses. We first ran DEC with no particular constraints, using only the distribution of extant species. We then performed DEC analyses including fossil information in the form of ‘fossil constraints’ at certain nodes, according to the range of distribution of fossil occurrences assigned to a particular taxon during the relevant time frame. For example the crown age of Carettochelyidae (Testudines) dates back to the Late Jurassic (150 Ma, node 5, Fig. 3; Supplementary Table 7), and we set a constraint on this node reflecting the distribution of all the Late Jurassic fossils attributed to Carettochelyidae. Similarly, for the origin of turtles (210 Ma, node 1), distribution constraints represent the range of Late Triassic fossils assigned to turtles. For the crown of Trionychidae, in the Early Cretaceous (123 Ma, node 2), the early fossils assigned to the clade were used to constrain the geographic origin of Trionychidae. In total, we implemented 23 fossil constraints for turtles (Supplementary Table 7), 30 fossil constraints for lepidosaurs (Supplementary Table 8), and 8 for crocodiles (Supplementary Table 9).
We included the fossil distribution in two different approaches: (i) a soft (SFC), and (ii) hard fossil constraints (HFC). For the SFC approach, fossil data were incorporated into the anagenetic component of the likelihood framework. The direct impact of a given fossil is limited to the particular branch to which it has been assigned, although it may indirectly influence other branches. The inclusion of a fossil conditions the estimated geographic-transition probability matrix for that branch by imposing a spatiotemporal constraint on the simulation process. Only the simulations resulting in a geographic range including the area of fossil occurrence contribute to the geographic-range transition probability matrix for the branch concerned; simulations not meeting this constraint are discarded112. For SFC, we used the command ‘fossil in DEC. We consider this to be a ‘soft’ constraint, because other areas different from that in which the fossil was found could be included in the ancestral states. In some cases, in which today’s diversity is not representative of past diversity (e.g. due to extreme levels of extinction), the SFC model may still overlook known fossil information. We therefore also implemented an HFC model in which the estimation of ancestral areas was fixed to the location of fossils. This was achieved with existing functions in the C++ version of Lagrange, using the command ‘fxnode’. By fixing nodes to the distribution area of fossils, we assume fossil occurrences reflect the distribution of the ancestors, i.e. that the fossil record is complete. This is a strong assumption, but it makes it possible to recover all fossil ranges in the ancestral estimations. The real scenario probably lies somewhere between the SFC and HFC inferences.
We then compared the timing and number of range extinction and dispersal events inferred with the three different biogeographic analyses. In DEC, range-subdivision (inheritance) scenarios (vicariance, duplication and peripatric isolation) occur at cladogenetic events, whereas extinction (range contraction) and dispersal (range expansion) are modeled as stochastic processes occurring along the branches of the tree113. As the probability of any extinction/dispersal event is constant along the entire length of the branch, we estimate the periods at which range extinction and dispersal occurred by dividing the phylogeny into intervals of 25 million years and calculating the number of branches for which extinction/dispersal was inferred crossing a particular time interval (the same branch could cross two continuous intervals).
Author contributions
Both authors designed the study, analyzed the data and wrote the manuscript.
Competing interests
The authors have no competing financial interests to declare.
Acknowledgements
The authors are very grateful to Arne Mooers, Joaquin Hortal, Juan Arroyo, and anonymous reviewers for comments and suggestions that greatly improved the study. Previous versions of the manuscript benefited from the comments of Gary Mittlebach, Emmanuelle Jousselin and Jonathan Rolland. Financial support was provided by a Marie-Curie FP7–COFUND (AgreenSkills fellowship–26719) grant to A.S.M. and a Marie Curie FP7-IOF (project 627684 BIOMME) grant to F.L.C. This work benefited from an “Investissements d’Avenir” grant managed by the “Agence Nationale de la Recherche” (CEBA, ref. ANR-10-LABX-25-01).
Footnotes
Data accessibility statement All the data used in this manuscript are presented in the manuscript and its supplementary material or have been published or archived elsewhere.
Reference list
- 1.↵
- 2.
- 3.
- 4.
- 5.
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.
- 46.↵
- 47.
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.
- 58.↵
- 59.↵
- 60.
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.
- 94.↵
- 95.↵
- 96.
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.↵
- 106.↵
- 107.↵
- 108.↵
- 109.↵
- 110.↵
- 111.↵
- 112.↵
- 113.↵